An Exploration of Machine Learning Models to Forecast the Unemployment Rate of South Africa: A Univariate Approach

Abstract

The South African unemployment rate is 29.1%, this is the highest unemployment rate that the country has recorded since the 1970s. The country is in the top ten countries with the highest unemployment rates in the world. COVID-19 threatens to increase the unemployment rate above the 50% mark. A public policy intervention is the most suitable instrument for the country in order to address this problem, however, policy is reliant on accurate and reliable forecasting. This paper explores univariate machine learning techniques to forecast the South African unemployment rate. Six traditional statistical models are compared with seven machine learning models. The multi-layer perceptron achieves the lowest error rate, whilst the ridge regression model achieved the highest R - squared. These are closely followed by ARIMA, LASSO, and the elastic net, showing that machine learning models can forecast the South African unemployment rate with higher accuracy than traditional statistical methods.